aistablev1.0.0

X-ray Pneumonia Detection

Flutter mobile app for chest X-ray pneumonia detection using deep learning. Supports VGG16, EfficientNet, InceptionV3, and ResNet50 with on-device TFLite inference.

flutterdeep-learningtensorflow-litemedical-imagingdart

View on GitHub

93.7%

best accuracy

95.4%

best F1 score

CNN architectures

on-device

TFLite inference

Sole engineer — model training + mobile app

The problem

Bringing chest X-ray pneumonia screening to low-resource settings means running deep-learning inference on-device, with no server round-trip or connectivity requirement.

Under the hood

Flutter App: Cross-platform mobile UI
Image Input: Camera capture or gallery pick
TFLite Runtime: On-device model inference
VGG16: 91.87% accuracy
EfficientNet: 90.13% accuracy
InceptionV3: 89.55% accuracy
ResNet50: 93.70% accuracy
Diagnosis: Confidence scores and classification

Outcome

ResNet50 reached 93.70% accuracy and 95.42% F1 on the held-out test set, running entirely on-device through TFLite — no image ever leaves the phone.

Models

93.7%

Best Accuracy

95.4%

Best F1 Score

Data Sent

X-ray Pneumonia Detection

TFLite Runtime
- On-device Inference
- Image Preprocessing
- Model Selector
runs models
Models
- VGG16 - 91.87%
- EfficientNet - 90.13%
- InceptionV3 - 89.55%
- ResNet50 - 93.70%
presents results
Flutter App
- Camera / Gallery Input
- Confidence Scores
- Model Comparison Dashboard

Four deep learning architectures: VGG16, EfficientNet B3, InceptionV3, and ResNet50 fine-tuned on chest X-ray dataset
On-device TensorFlow Lite inference - all processing runs locally, no data leaves the device
ResNet50 leads with 93.70% accuracy and 95.42% F1 score on the test set
Real-time image analysis with detailed confidence scores for transparent decision-making
Built-in model comparison dashboard showing accuracy and F1 metrics side by side
Modern Flutter UI with gradient design, smooth animations, and cross-platform support
Camera capture and gallery import for flexible X-ray image input
Minimum Android SDK 21 / iOS 12.0 - runs on 4+ year old devices

Want systems built like this?

I'm open to backend and platform work, contract or permanent — on-site in Belgium and the Netherlands, or fully remote for a team anywhere in the world, US and APAC hours included. EU citizen, so no sponsorship is needed. Tell me what you're building.

Download CV

X-ray Pneumonia Detection

The problem

Under the hood

Outcome

TFLite Runtime

Models

Flutter App

Want systems built like this?

TFLite Runtime

Models

Flutter App